Having the ability to predict the course of future events is an attractive idea, but difficult to achieve in practice because of the vast number of possibilities. This diploma presents an attempt at doing so based on the hypothesis that patterns of events from the past tend to repeat. We model real world events as groups of concepts. We cluster events into groups of related events. We then build a dataset from these clusters, where the attributes of each data point represent concepts that occur in a single cluster within a certain time window, and the target labels are concepts that occur the next week. We compare the effectiveness of several different prediction models. Our tests show that relationships between concepts contain useful information for predicting the future course of events.
|
